SDROC Estimate ROC of a classifierR=SDROC(TS,P) % provide test set and trained classifier R=SDROC(OUT) % provide soft outputs out=ts*-p R2=SDROC(R,OUT) % re-estimate from OUT using op.points in R INPUT TS test set P trained classifier pipeline OUT data set with classifier soft outputs OUTPUT R,R2 SDROC objects OPTIONS 'target' Name of the target decision 'non-target' Name of the non-target decision in the resulting op.point. 'reject' Add a reject option and construct reject curve. - if (0,1) fraction is given, set threshold by rejecting percentage of all samples - if SDOPS or SDROC is given, use current op.point 'measures' Cell array with measure names and parameters. 'noconfmat' - Do not store confusion matrices ('confmat' option is used by default) 'polarity',P - Set polarity of the soft output (P is 'similarity' or 'distance') 'maxpoints',M - Set maximum number of operating point (2000 by default) DESCRIPTION SDROC performs ROC analysis of classifier P on the test set TS. Alternatively, soft output set OUT can be provided. SDROC performs two- or multi- class analysis using output thresholding or weighting. The output is an object with estimated measures as a set of operating points. ROC can be visualized using interactive SDDRAWROC plot. Operating points can be selected using SETCUROP or CONSTRAIN. EXAMPLES Estimate from test set and trained classifier: p=sdparzen(tr) r=sdroc(ts,p) Estimate from soft outputs out=ts*-p % -p removes decision step so that out is sddata with soft outputs r=sdroc(out) Specifying the performance measures to estimate: r=sdroc(ts,p,'measures',{'FPr','apple','TPr','apple'}) r=sdroc(ts,p,'measures',{'custom:F',@custom_F_measure}) % custom measure SEE ALSO CUSTOM_F_MEASURE, SETCUROP